In the 2020 General Election, using the first round of absentee data, the KDP Data Team was able to build out a predictive model to help campaigns identify and target voters who were more likely to request an absentee ballot. For more information on that model, please visit: https://docs.google.com/document/d/1U6WMT851QMvVtfn5mP0QEPqwnSxWptF-5KZ64B-sJR0/edit?usp=sharing..
During the cycle, internal documentation was created to determine how the KDP model was performing, and compare it against another common vote-by-mail (VBM) model created by Civis.
As demonstrated by Figure 1 below, the KDP model scored a wider range of the Kentucky electorate, and was more accurate than the Civis model within the score buckets (0-9.99, 10-19.99, etc).
When looking at the individual scores (Figure 2), this visualization confirms that the initial version of the KDP Model (KDP Version 1) was the best performing model when compared to how a perfect model (the straight redline from bottom left to top right). The points on the scatter plot indicate the total percentage of voters who did request ballots in 2020. As the graph shows, KDP Version 1 fits the line better than all the other models.
Due to the performance of this model in 2020, the KDP Ballot Request Model has been tweaked to be an Absentee and Early Vote model going forward. Necessary data points will be switched out to best reflect the specific election they are built for (for instance, switching out the 2020 turnout models for 2022 turnout models).
| FIGURE 1 | FIGURE 2 |
 |  |